1
00:00:00,140 --> 00:00:02,839
Welcome back, brought to you by Jellypod
AI!

2
00:00:03,439 --> 00:00:06,659
You know, if you spent any time running
background agents in Claude Code,

3
00:00:06,839 --> 00:00:07,639
you know the pain.

4
00:00:08,019 --> 00:00:10,719
You hit control C or send a prompt to the
background, and then...

5
00:00:12,420 --> 00:00:16,979
to manage it, you had to navigate through
these interactive TUI menus.

6
00:00:17,020 --> 00:00:20,760
It just felt clunky if you wanted to
quickly script things or check up from a shell

7
00:00:20,819 --> 00:00:21,079
prompt.

8
00:00:21,224 --> 00:00:22,584
Oh, mate, spot on.

9
00:00:22,944 --> 00:00:25,324
It was like trying to herd wombats in the
dark.

10
00:00:25,864 --> 00:00:28,284
You just wanted to run a simple command
from your terminal,

11
00:00:28,664 --> 00:00:31,524
but instead you were stuck poking around a
full screen interface.

12
00:00:32,005 --> 00:00:36,244
But, look, version 2.1.251 completely
fixes that.

13
00:00:36,804 --> 00:00:40,524
They brought direct Unix style subcommands
right into the main CLI.

14
00:00:40,782 --> 00:00:45,402
Right, so according to the documentation,
Anthropic added five key subcommands

15
00:00:45,442 --> 00:00:46,362
straight to claude help.

16
00:00:46,862 --> 00:00:51,262
You have claude attach, claude logs,
claude stop, claude respawn,

17
00:00:51,442 --> 00:00:52,522
and claude rm.

18
00:00:53,003 --> 00:00:55,843
It turns background sessions into first
class CLI primitives.

19
00:00:55,994 --> 00:00:56,934
Yeah, yeah, exactly.

20
00:00:58,214 --> 00:01:02,335
Think of it like managing Docker
containers or background system services,

21
00:01:02,414 --> 00:01:02,634
right?

22
00:01:03,494 --> 00:01:07,175
So if you run a session in the background
with claude flag flag bg,

23
00:01:07,974 --> 00:01:12,694
say for running a full test suite, you do
not have to pull up the whole agent UI

24
00:01:12,734 --> 00:01:13,114
anymore.

25
00:01:13,286 --> 00:01:16,286
Walk me through what that looks like in a
real workflow.

26
00:01:16,407 --> 00:01:21,967
Say I start a heavy background job like
claude flag flag bg run full test suite and

27
00:01:21,986 --> 00:01:23,026
fix failing mocks.

28
00:01:23,587 --> 00:01:24,226
What do I do next?

29
00:01:24,292 --> 00:01:27,732
Right, so you kick that off, and it gives
you a session ID.

30
00:01:27,912 --> 00:01:32,852
Now, instead of jumping into a big menu,
you can split your tmux window or open

31
00:01:32,892 --> 00:01:37,492
another terminal tab and just type claude
logs followed by that session ID.

32
00:01:37,705 --> 00:01:42,052
It streams the output right there in real
time, like tailing a log file.

33
00:01:42,141 --> 00:01:45,042
That is huge for automation and terminal
splits!

34
00:01:45,542 --> 00:01:49,421
And if you see something in the logs where
the agent needs your input or gets stuck,

35
00:01:49,941 --> 00:01:54,121
you can just run claude attach with that
session ID, hop right into the interactive

36
00:01:54,241 --> 00:01:56,721
session, answer the prompt, and keep
moving!

37
00:01:57,329 --> 00:01:57,850
Bang on.

38
00:01:58,509 --> 00:02:01,369
And if it goes off the rails, claude stop
kills it.

39
00:02:02,109 --> 00:02:05,170
If you need to restart it with all its
prior context intact,

40
00:02:05,630 --> 00:02:07,869
claude respawn brings it right back to
life.

41
00:02:08,349 --> 00:02:13,129
Then when you are all done, claude rm
cleans up the transcript and session storage

42
00:02:13,510 --> 00:02:15,749
so your disk does not end up full of old
garbage.

43
00:02:15,875 --> 00:02:21,795
Now, there were a couple of really crucial
bug fixes around this in versions 2.1.251

44
00:02:21,935 --> 00:02:25,955
and 2.1.252 that people should know about,
right?

45
00:02:26,117 --> 00:02:26,877
Oh, absolutely.

46
00:02:27,178 --> 00:02:31,977
In 2.1.251, they fixed a really annoying
bug where background sessions and their

47
00:02:32,037 --> 00:02:37,177
subagents could not edit files inside a
git worktree created with git worktree add.

48
00:02:37,777 --> 00:02:41,177
That was a big bottleneck if your
background worker was trying to isolate its code

49
00:02:41,237 --> 00:02:43,077
changes in a separate branch worktree.

50
00:02:43,789 --> 00:02:46,149
Yeah, that would completely block
autonomous workflows.

51
00:02:46,789 --> 00:02:53,269
And then in version 2.1.252, which just
landed, they patched a Mac specific bug.

52
00:02:53,349 --> 00:02:57,829
Some macOS users were seeing Bash commands
fail with an error saying task output

53
00:02:57,949 --> 00:03:01,069
swap refused, tasks dir moved or linked.

54
00:03:01,083 --> 00:03:05,802
Yeah, it happened when your tasks
directory was symlinked or moved around on macOS.

55
00:03:05,803 --> 00:03:12,443
2.1.252 sort of untangled that symlink
swap check, so Mac devs can run background

56
00:03:12,536 --> 00:03:14,283
tasks without hitting that wall.

57
00:03:15,057 --> 00:03:17,357
What else landed in 2.1.252?

58
00:03:18,137 --> 00:03:20,956
I saw a couple of nice quality of life
tweaks in the changelog.

59
00:03:21,000 --> 00:03:23,640
Well, one big one is for fresh
repositories.

60
00:03:23,771 --> 00:03:29,960
Previously, if your project did not have a
dot claude slash settings dot local dot

61
00:03:30,088 --> 00:03:35,160
json file yet, clicking always allow on
permission prompts would fail to save.

62
00:03:35,380 --> 00:03:42,200
Now, 2.1.252 creates that file on the fly
and saves your preference properly.

63
00:03:42,208 --> 00:03:43,408
That is super clean.

64
00:03:43,528 --> 00:03:48,608
And they also fixed an issue where
background task notifications with huge failure

65
00:03:48,665 --> 00:03:53,328
output, like a massive git error when a
disk is full, were bloating the session

66
00:03:53,379 --> 00:03:56,688
transcript and making it crash into the
API request size limit!

67
00:03:57,347 --> 00:03:57,727
Right!

68
00:03:58,387 --> 00:04:03,507
Imagine your disk filling up, and the
error log itself is so huge it blows past the

69
00:04:03,607 --> 00:04:04,747
API payload limit.

70
00:04:05,628 --> 00:04:07,208
Talk about adding insult to injury!

71
00:04:07,827 --> 00:04:09,547
But yeah, it truncates gracefully now.

72
00:04:09,662 --> 00:04:13,983
Honestly, having native terminal
subcommands like logs and attach makes scripting

73
00:04:14,003 --> 00:04:17,242
local AI workflows feel so much more
natural.

74
00:04:17,282 --> 00:04:21,023
You can actually write bash scripts or
shell aliases around your background agents

75
00:04:21,102 --> 00:04:21,282
now.

76
00:04:21,826 --> 00:04:26,366
Oh, mate, I am already writing little
watchdog scripts in my shed setup to tail

77
00:04:26,547 --> 00:04:29,966
agent logs and notify my phone when a
build finishes.

78
00:04:30,026 --> 00:04:34,946
It is proper Unix philosophy applied to AI
agents, and it makes life so much easier.

79
00:04:35,042 --> 00:04:36,162
That is fantastic.

80
00:04:36,242 --> 00:04:42,962
Definitely update to Claude Code 2.1.252
to get all these CLI commands and fixes.

81
00:04:43,065 --> 00:04:46,082
Thanks for listening, and we will see you
next time!